// CHECK-NEXT:      %18 = "tfl.pseudo_const"() <{value = dense<2> : tensor<1xi32>}> : () -> tensor<1xi32>
// CHECK-NEXT:      %19 = "tfl.concatenation"(%18, %17, %16) <{axis = 0 : i32, fused_activation_function = "NONE"}> : (tensor<1xi32>, tensor<1xi32>, tensor<1xi32>) -> tensor<3xi32>
// CHECK-NEXT:      %20 = mhlo.dynamic_reshape %2, %19 : (tensor<2x3x4x?xf32>, tensor<3xi32>) -> tensor<2x3x?xf32>

  %min = arith.constant dense<0.0> : tensor<16xf32>
  %max = arith.constant dense<255.0> : tensor<16xf32>
  %mini = "tf.Identity"(%min) : (tensor<16xf32>) -> tensor<16xf32>
  %maxi = "tf.Identity"(%max) : (tensor<16xf32>) -> tensor<16xf32>
  %fq = "tf.FakeQuantWithMinMaxVarsPerChannel"(%in, %mini, %maxi) {num_bits = 5, narrow_range = false} : (tensor<3x3x3x16xf32>, tensor<16xf32>, tensor<16xf32>) -> tensor<3x3x3x16xf32>

    func.func @main() -> tensor<1xi32> {
      %0 = "tf.BadValue"() {value = dense<1000> : tensor<1xi32>} : () -> tensor<1xi32>
      %1 = "tf.AlsoBad"() {value = dense<10> : tensor<1xi32>} : () -> tensor<1xi32>
      func.return %0 : tensor<1xi32>
    }
  })";
  CellReader<int64_t> error(kFailedLegalizationStreamz);
  CreateModule(kMultipleFailures);

    %2 = stablehlo.reshape %1 : (tensor<i32>) -> tensor<1xi32>
    %3 = stablehlo.concatenate %2, %cst_0, %cst_1, %cst_2, dim = 0 : (tensor<1xi32>, tensor<1xi32>, tensor<1xi32>, tensor<1xi32>) -> tensor<4xi32>
    %4 = stablehlo.dynamic_broadcast_in_dim %arg2, %3, dims = [0, 1, 2, 3] : (tensor<1x1x1x2xf32>, tensor<4xi32>) -> tensor<?x3x4x2xf32>
    %5 = stablehlo.add %0, %4 : tensor<?x3x4x2xf32>

  %min = arith.constant dense<0.0> : tensor<16xf32>
  %max = arith.constant dense<15.0> : tensor<16xf32>
  %mini = "tf.Identity"(%min) : (tensor<16xf32>) -> tensor<16xf32>
  %maxi = "tf.Identity"(%max) : (tensor<16xf32>) -> tensor<16xf32>
  %fq = "tf.FakeQuantWithMinMaxVarsPerChannel"(%in, %mini, %maxi) {num_bits = 3, narrow_range = false} : (tensor<3x3x3x16xf32>, tensor<16xf32>, tensor<16xf32>) -> tensor<3x3x3x16xf32>

  %min = arith.constant dense<0.0> : tensor<16xf32>
  %max = arith.constant dense<15.0> : tensor<16xf32>
  %mini = "tf.Identity"(%min) : (tensor<16xf32>) -> tensor<16xf32>
  %maxi = "tf.Identity"(%max) : (tensor<16xf32>) -> tensor<16xf32>
  %fq = "tf.FakeQuantWithMinMaxVarsPerChannel"(%in, %mini, %maxi) {num_bits = 3, narrow_range = false} : (tensor<3x3x3x16xf32>, tensor<16xf32>, tensor<16xf32>) -> tensor<3x3x3x16xf32>

    // CHECK:     return %[[RESULT]]
    func.return %0 : tensor<10xi32>
  }

  // CHECK-LABEL: func @foo
  // CHECK-SAME:    %[[ARG0:.*]]: tensor<10xi32>, %[[ARG1:.*]]: tensor<10xi32>
  func.func @foo(%arg0: tensor<10xi32>, %arg1: tensor<10xi32>) -> tensor<10xi32> {

    // CHECK: return %[[RESULT]]
    func.return %0 : tensor<10xi32>
  }

  // CHECK-LABEL: func private @_tf_func
  func.func private @_tf_func(%arg0: tensor<?xi32>, %arg1: tensor<*xi32>) {
    // CHECK: tf.StreamResults

    // StreamResults is a pseudo op in this test.

func.func @xla_gather_known_output_shape(%arg0: tensor<5xi32>, %arg1: tensor<1xi32>, %arg2: tensor<1xi32>) -> tensor<i32> {
  // dimension_numbers: {
  //   collapsed_slice_dims: 0
  //   start_index_map: 0
  // }
  %0 = "tf.XlaGather"(%arg0, %arg1, %arg2) {device = "", dimension_numbers = "\12\01\00\1A\01\00", indices_are_sorted = true} : (tensor<5xi32>, tensor<1xi32>, tensor<1xi32>) -> tensor<i32>

  %119 = "tf.StridedSlice"(%116, %16, %15, %16) {begin_mask = 0 : i64, device = "", ellipsis_mask = 0 : i64, end_mask = 1 : i64, new_axis_mask = 0 : i64, shrink_axis_mask = 0 : i64} : (tensor<1xi32>, tensor<1xi32>, tensor<1xi32>, tensor<1xi32>) -> tensor<0xi32>